A method, system, device and medium for insulator contamination risk assessment and prediction

Abstract

The application discloses a kind of insulator contamination risk assessment and prediction method, system, equipment and medium, belong to power system state assessment and intelligent operation technology field, this method includes: collecting and pre-processing the multi-source data related to outdoor insulator contamination risk, construct time-space alignment multidimensional time series feature vector;For each target node, construct continuous contamination risk index;According to the action relationship between pollution source emission, meteorological condition, factory layout and insulator position, construct the causal prior information for depicting action relationship;Multi-dimensional time series feature vector and corresponding continuous contamination risk index are used to train the spatiotemporal prediction model established in advance, and the insulator contamination risk prediction model is obtained.The application can more fully utilize the multi-source data related to outdoor insulator risk, construct the contamination risk index with clear physical meaning, and introduce mechanism-driven causal prior constraint, improve the accuracy and reliability of insulator contamination risk prediction.

Description

Technical Field

[0001] This application belongs to the field of power system condition assessment and intelligent operation and maintenance technology, specifically involving a method, system, equipment and medium for assessing and predicting insulator pollution risk. Background Technology

[0002] Outdoor high-voltage insulators are widely used in substations and transmission lines, and are exposed to complex atmospheric environments for extended periods. Affected by industrial emissions, dust, sea salt, humidity, and precipitation, a contamination layer easily forms on their surface. Under suitable meteorological conditions, this contamination layer absorbs moisture or dissolves, leading to a significant increase in leakage current and a decrease in insulation strength. In severe cases, this can cause flashover faults, resulting in equipment tripping and power outages. Therefore, conducting pollution risk assessments and predictions for outdoor insulators is a crucial foundation for ensuring the safe operation of the power grid and implementing risk-based condition-based maintenance.

[0003] In existing technologies, a common approach involves manually sampling and laboratory testing to measure pollution levels such as equivalent salt density (ESDD) and non-soluble deposition density (NSDD), and then combining this with relevant standards or empirical curves to conduct offline assessments of pollution levels and flashover voltage. While this method has played a role in insulation design and long-term trend analysis, it suffers from long sampling cycles, heavy workloads, and difficulty in reflecting rapid changes in on-site pollution and weather conditions, thus failing to meet the needs of online continuous risk assessment. With the development of online monitoring technology, some substations have installed leakage current monitoring devices on insulators at typical locations and are attempting to predict leakage current changes using moving averages, ARIMA, grey prediction models, and time series models based on neural networks (such as BP neural networks and LSTM), or combining simple thresholds for early warning. These methods mainly suffer from the following problems:

[0004] (1) Most methods only treat leakage current as a single time series for modeling, failing to make full use of multi-source information such as pollution source emissions, meteorological conditions, and environmental background, and not giving enough consideration to the key driving factors in the formation and evolution of pollution.

[0005] (2) The prediction targets are usually leakage current values ​​or discrete pollution levels. There is a lack of continuous pollution risk indicators with clear physical meaning that can be compared between different sites and operating conditions, which is not conducive to carrying out risk assessment and maintenance decisions based on a unified scale.

[0006] (3) The model parameters and alarm thresholds are mostly based on experience and are sensitive to seasonal changes, changes in operating conditions, etc. The model has limited stability and generalization ability and it is difficult to provide reliable risk assessment results in complex operating environments.

[0007] In summary, existing insulator pollution risk assessment methods based on traditional measurement and time-series prediction still have shortcomings in terms of multi-source information fusion, continuous risk characterization, and prediction reliability, making it difficult to provide accurate and stable risk assessment results continuously in complex operating environments. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this application proposes a method, system, device, and medium for assessing and predicting insulator pollution risk. This method can make fuller use of multi-source data related to outdoor insulator risk, construct pollution risk indicators with clear physical meaning, and introduce mechanism-driven causal prior constraints to improve the accuracy and reliability of insulator pollution risk prediction.

[0009] This application is achieved through the following technical solution:

[0010] A method for assessing and predicting insulator pollution risk includes:

[0011] Multi-source data related to the pollution risk of outdoor insulators are collected and preprocessed to construct a spatiotemporally aligned multi-dimensional time-series feature vector; the multi-source data includes insulator status data, pollution source emission data, and meteorological environmental data.

[0012] Based on the multidimensional temporal feature vector, a continuous pollution risk index is constructed for each target node, serving as a label for the pollution risk intensity of the corresponding target node at each time.

[0013] Based on the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location, causal prior information is constructed to characterize the interaction.

[0014] The pre-established spatiotemporal prediction model is trained using the multidimensional temporal feature vector and the corresponding continuous pollution risk index to obtain an insulator pollution risk prediction model. The spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of the causal prior information. The temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk index at the prediction time. During training, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms.

[0015] New multi-source data related to the pollution risk of outdoor insulators are collected in real time and preprocessed to update the multi-dimensional time-series feature vector. The updated multi-dimensional time-series feature vector is then input into the insulator pollution risk prediction model, and the insulator pollution risk prediction result is output.

[0016] In some implementations, the multidimensional time-series feature vector includes: long-term pollution accumulation features, short-term excitation features, and pollution source-related features;

[0017] The long-term pollution accumulation feature represents the degree of long-term pollution accumulation and is extracted from the insulator condition data.

[0018] The short-term excitation feature represents the characteristics of short-term excitation conditions and is extracted from the meteorological environment data.

[0019] The pollution source characteristics represent the characteristics of the impact of upstream pollution sources, which are extracted from the pollution source emission data.

[0020] In some implementations, the continuous pollution risk index is constructed as follows:

[0021] By weighting and normalizing the long-term pollution accumulation characteristics, short-term excitation characteristics, and pollution source-related characteristics, a continuous pollution risk index is obtained.

[0022] In some implementations, the process of constructing the causal prior information includes:

[0023] Constructing node types and node sets: Abstracting relevant entities in the scene into nodes of different types to form node sets, and associating each node in the node set with its node type, including substation / insulator node type, pollution source node type and meteorological environment node type;

[0024] Constructing Relationship Types: Based on physical mechanisms and engineering experience, several types of relationships between nodes are defined to form a relationship set. The relationship set includes pollutant transport relationships, environmental similarity relationships, and plant topology or electrical connection relationships.

[0025] Construct a structure mask matrix: For each relation type in the relation set, construct a structure mask matrix. The elements in the structure mask matrix are used to indicate whether a directed edge from one node to another is allowed under the corresponding relation type. If it exists, the element is set to 1; otherwise, it is set to 0.

[0026] Constructing edge weight prior information: Under the premise of structural masking, edge weight priors are defined based on the distance between nodes, wind direction and speed, emission intensity and meteorological similarity. These priors are used to characterize the relative influence of one node on another node in the corresponding relationship type, and the edge weight priors are normalized.

[0027] In some implementations, the construction and training process of the insulator pollution risk prediction model includes:

[0028] The multidimensional temporal feature vector is input into the node feature encoding part, which maps the temporal feature vectors of the input heterogeneous nodes to a latent space of a unified dimension and outputs the node encoded representation to the spatial dependency modeling part.

[0029] The spatial dependency modeling part constructs a multi-relationship graph structure based on the causal prior information, and models the spatial dependency between nodes through a graph neural network to obtain an updated node spatial encoding representation.

[0030] The node spatial encoding representation is input into the time dependency modeling part according to a preset time window. The time dependency modeling part performs time modeling and outputs the comprehensive spatiotemporal representation of the node.

[0031] By inputting the comprehensive spatiotemporal representation of the nodes into the fully connected network, the predicted value of the continuous pollution risk index at the target prediction time is obtained.

[0032] In some implementations, the spatial dependency modeling part uses a graph neural network with an attention mechanism for spatial encoding, establishes propagation connections only between node pairs allowed by the structural mask, and performs weighted aggregation of information propagation between nodes based on edge weight priors; through several layers of graph neural networks, a node spatial encoding representation that integrates spatial dependency information is obtained.

[0033] In some implementations, the constraint term is constructed by calculating the deviation between the attention weights between nodes learned by the model and the prior information of edge weights, and using this deviation as part of the total loss function to guide the edge influence strength learned by the model to be consistent with the causal prior information.

[0034] Secondly, this application proposes an insulator pollution risk assessment and prediction system, including:

[0035] The data acquisition and preprocessing unit is configured to: acquire multi-source data related to the pollution risk of outdoor insulators and preprocess it to construct a spatiotemporally aligned multi-dimensional time-series feature vector; the multi-source data includes insulator status data, pollution source emission data and meteorological environment data.

[0036] The indicator construction unit is configured to: construct a continuous pollution risk indicator for each target node based on the multi-dimensional time-series feature vector, which serves as a label for the pollution risk intensity of the corresponding target node at each time.

[0037] The causal prior construction unit is configured to: construct causal prior information to characterize the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location;

[0038] The model training and update unit is configured to: train a pre-established spatiotemporal prediction model using the multi-dimensional temporal feature vector and the corresponding continuous pollution risk index to obtain an insulator pollution risk prediction model; wherein the spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of the causal prior information. The temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk index at the prediction time. During the training process, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function containing prediction error terms, constraint terms, and parameter regularization terms.

[0039] Furthermore, the prediction unit is configured to: acquire new multi-source data in real time using the data acquisition and preprocessing unit and perform preprocessing, update the multi-dimensional time-series feature vector, input the updated multi-dimensional time-series feature vector into the insulator pollution risk prediction model, and output the insulator pollution risk prediction result.

[0040] Thirdly, this application proposes an electronic device including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the above-described methods for assessing and predicting pollution risks of insulators.

[0041] Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for assessing and predicting pollution risks of insulators.

[0042] This application proposes a method for assessing and predicting pollution risks of insulators, which can fully integrate multi-source data related to the risks of outdoor insulators and introduce mechanism-driven causal prior constraints to improve the accuracy, reliability and interpretability of insulator pollution risk prediction results, and provide reliable quantitative support for risk condition maintenance of outdoor insulators.

[0043] Accordingly, the insulator pollution risk assessment and prediction system, electronic device and computer-readable storage medium proposed in this application also have the same technical effects as described above. Attached Figure Description

[0044] The accompanying drawings, which are included to provide a further understanding of the embodiments of this application and form part of this application, do not constitute a limitation on the embodiments of this application. In the drawings:

[0045] Figure 1 This is a flowchart of the insulator pollution risk assessment and prediction method proposed in the embodiments of this application;

[0046] Figure 2 This is a schematic diagram of the spatiotemporal prediction model structure according to an embodiment of this application;

[0047] Figure 3 This is a block diagram illustrating the principle of the insulator pollution risk assessment and prediction system proposed in this application.

[0048] Figure 4 This is a schematic diagram of the electronic device proposed in the embodiments of this application;

[0049] Figure 5 This is a schematic diagram of a computer-readable storage medium proposed in an embodiment of this application.

[0050] Figure reference numerals and corresponding component names:

[0051] 300 - Prediction system; 301 - Data acquisition and preprocessing unit; 302 - Indicator construction unit; 303 - Causal prior construction unit; 304 - Model training and update unit; 305 - Prediction unit; 400 - Electronic device; 410 - Memory; 420 - Processor; 411 - Computer program A; 500 - Computer-readable storage medium; 511 - Computer program B. Detailed Implementation

[0052] In the following, the terms “comprising” or “may include” as used in the various embodiments of this application indicate the presence of a function, operation, or element of the invention and do not limit the addition of one or more functions, operations, or elements. Furthermore, as used in the various embodiments of this application, the terms “comprising,” “having,” and their cognates are intended only to indicate a specific feature, number, step, operation, element, component, or combination of the foregoing and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations of the foregoing, or adding one or more combinations of the foregoing.

[0053] In various embodiments of this application, the expression "or" or "at least one of A and / or B" includes any combination or all combinations of the words listed simultaneously. For example, the expression "A or B" or "at least one of A and / or B" may include A, may include B, or may include both A and B.

[0054] The terms used in the various embodiments of this application (such as "first," "second," etc.) may modify various constituent elements in the various embodiments, but do not limit the corresponding constituent elements. For example, the above terms do not limit the order and / or importance of the elements. The above terms are only used for the purpose of distinguishing one element from other elements. For example, a first user device and a second user device refer to different user devices, although both are user devices. For example, without departing from the scope of the various embodiments of this application, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element.

[0055] It should be noted that if a description is made of "connecting" one component to another, then the first component can be directly connected to the second component, and a third component can be "connected" between the first and second components. Conversely, when a component is "directly connected" to another component, it can be understood that there is no third component between the first and second components.

[0056] The terminology used in the various embodiments of this application is for the purpose of describing particular embodiments only and is not intended to limit the various embodiments of this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. The terms (such as those defined in a generally used dictionary) are to be interpreted as having the same meaning as in the context of the relevant technical field and are not to be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the embodiments and accompanying drawings. The illustrative embodiments and descriptions of this application are only for explaining this application and are not intended to limit this application.

[0058] like Figure 1 As shown in the embodiments of this application, a method for assessing and predicting insulator pollution risk is proposed, including the following steps:

[0059] Step 1: Collect multi-source data related to the pollution risk of outdoor insulators and preprocess the data to construct a spatiotemporally aligned multi-dimensional time-series feature vector; the multi-source data includes insulator status data, pollution source emission data, and meteorological environmental data.

[0060] Step 2: Based on the multi-dimensional temporal feature vector, construct a continuous pollution risk index for each target node, which serves as the pollution risk intensity label for the corresponding target node at each time.

[0061] Step 3: Based on the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location, construct causal prior information to characterize the interaction.

[0062] Step 4: Train the pre-established spatiotemporal prediction model using multi-dimensional temporal feature vectors and corresponding continuous pollution risk indicators to obtain the insulator pollution risk prediction model. This spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of causal prior information, while the temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk indicator at the prediction time. During the training process, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms.

[0063] Step 5: Collect new multi-source data related to the pollution risk of outdoor insulators in real time and preprocess it to update the multi-dimensional time series feature vector. Then, input the updated multi-dimensional time series feature vector into the insulator pollution risk prediction model and output the insulator pollution risk prediction result.

[0064] Furthermore, in step 1 of this application embodiment, the multi-source data collected includes, but is not limited to, the following categories:

[0065] (1) Insulator condition data, including online monitoring data of leakage current of outdoor insulators (sampling interval can be minute or hour), outdoor insulator ESDD (equivalent salt density) / NSDD (non-soluble deposition density) measurement data carried out periodically or as needed, and at least one of the following: insulator defect records, pollution flashover event records, etc.

[0066] (2) Pollution source emission data, including continuous monitoring data of smoke / particulate matter emissions from the chimney outlet (emission concentration, flue gas flow rate, etc.) and at least one of the following data: unit output, load curve, start-up and shutdown records, and desulfurization and denitrification device operation status;

[0067] (3) Meteorological environmental data, including at least one of the meteorological monitoring data such as wind direction, wind speed, temperature, relative humidity, precipitation, and atmospheric pressure. If necessary, regional air quality monitoring data or reanalysis data may also be included.

[0068] The data preprocessing process includes:

[0069] Time alignment and interpolation processing: Select a uniform time step to align data with different sampling frequencies to a uniform time axis, and fill missing points with forward padding, linear interpolation or other methods;

[0070] Outlier detection and removal: Identifying and removing values ​​that clearly exceed the physical range or are abnormal from sensors must be done using methods such as reasonable thresholds, box plots, or experience-based rules.

[0071] Normalization and standardization are performed to facilitate subsequent model training. Features of different dimensions are normalized or standardized according to node type. For example, continuous features are standardized according to mean or variance.

[0072] Through the above preprocessing, time-series feature vectors of various nodes are obtained, including long-term pollution accumulation features, short-term excitation features, and pollution source-related features. Among them, the long-term pollution accumulation features represent the characteristics of the long-term pollution accumulation degree, which are extracted from the insulator status data and include at least one of ESDD / NSDD and long-term leakage current within a certain time window; the short-term excitation features represent the characteristics of short-term excitation conditions, which are extracted from meteorological environmental data and include at least one of meteorological information such as relative humidity, precipitation, and fog in the recent period (e.g., within the last few hours); the pollution source-related features represent the characteristics of the influence of upstream pollution sources, which are extracted from pollution source emission data and include at least one of smoke and dust emission intensity, unit load level, and wind direction and speed in the corresponding period.

[0073] The multi-source data collected in this step also includes: plant topology and spatial location information, including: spatial coordinates of substations, outdoor insulators, pollution sources and meteorological monitoring points; and information on obstacles such as major buildings, pipe corridors, and mountains that have a significant impact on the local wind field; the plant topology and spatial location information is processed by coordinate normalization, obstacle information classification and spatial relationship structuring, etc., to provide a unified standard of data input for the subsequent construction of causal prior information.

[0074] Furthermore, in step 2 of this application embodiment, the method for constructing the continuous pollution risk indicator includes:

[0075] Based on long-term pollution accumulation characteristics, short-term excitation characteristics, and pollution source-related characteristics, a continuous pollution risk index is constructed for each insulator or substation node (i.e., target node), serving as a label for the pollution risk intensity of that node at each time point. Specifically, weighted summation, normalized nonlinear transformation, and other methods can be used to combine the various sub-features into a continuous pollution risk index with a unified scale, expressed as:

[0076] ;

[0077] in, , , These respectively represent the characteristics of long-term pollution accumulation, short-term triggering characteristics, and pollution source-related characteristics; The mapping function can be represented by weighted summation and normalized nonlinear transformation. This represents a continuous pollution risk index. That is, the final obtained continuous pollution risk index falls within a preset range, for example... or This facilitates comparisons between different devices and time periods. The continuous pollution risk index obtained in this step serves as the target output value for model training and evaluation.

[0078] Furthermore, in step 3 of this embodiment, based on the interaction between pollution source emissions, meteorological conditions, plant layout, and insulator locations, causal prior information is established to characterize the interaction between "pollution source-environment-equipment," which is used to constrain the information propagation paths and relative weights between nodes in the subsequent model. The specific construction process of the causal prior information includes:

[0079] Constructing node types and node collections: Abstracting relevant entities in the scene into different types of nodes to form node collections. This includes substation / insulator node sets. Pollution source node set and meteorological environment node set The substation / insulator node set represents the target equipment for which pollution risk needs to be predicted; the pollution source node set represents major emission sources such as boilers and chimneys; and the meteorological environment node set represents meteorological monitoring points or virtual nodes representing the environmental characteristics of a certain area. Each node... Associate its node type ,For example Where S represents the substation / insulator node type, E represents the pollution source node type, and M represents the meteorological environment node type.

[0080] Constructing Relationship Types: Based on physical mechanisms and engineering experience, define several types of relationships between nodes to form a relationship set. This includes: pollutant transport relationships, used to describe the impact of pollution sources on target insulators or substations under certain wind direction or distance conditions; environmental similarity relationships, used to describe the degree of similarity of meteorological conditions or environmental states at different nodes; and plant topology or electrical connection relationships, used to describe the topological associations between equipment.

[0081] Construct a structure mask matrix: for each relation type Construct the structure mask matrix Elements in the structure mask matrix (i.e., structure mask) is used to represent relation types Is it allowed for slave nodes to exist? Pointing to node Directed edges (i.e., nodes) For nodes (Whether there is an impact). Specifically: when judging the relationship type based on wind direction, distance, terrain, etc. Next node For nodes When there is a potential impact, set When the influence in this direction is considered negligible or inconsistent with the mechanism, set... .

[0082] Constructing edge weight prior information: Provided the structure mask allows (i.e., structure mask is 1), edge weight priors can be defined based on factors such as distance between nodes, wind direction and speed, emission intensity, and meteorological similarity. Used to characterize relation types Next node For nodes The relative influence intensity. Preferably, it can be applied to nodes of the same relation type that point to the same target node. After normalizing the edges, we get:

[0083] ;

[0084] in, For relation types Next node For nodes The a priori normalized value of the edge weight; For relation types Down pointer node Priorities of edge weights; To prevent tiny constants with a denominator of zero.

[0085] The structural mask matrix and edge weight prior information (normalized values) constructed in this step are used as causal prior constraints in subsequent models.

[0086] Furthermore, such as Figure 2 As shown, in step 4 of this embodiment, the construction and training process of the insulator pollution risk prediction model includes:

[0087] The temporal feature vectors (i.e., multidimensional temporal feature vectors) of each node are input into the spatiotemporal prediction model, which includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part.

[0088] The node feature encoding part maps the temporal feature vectors of the input heterogeneous nodes to a latent space of uniform dimension; specifically, it maps the original features of node v at time t. According to its node type Choose the appropriate encoding mapping and map it to a latent space of uniform dimension. For example:

[0089] ;

[0090] in, , These are the trainable parameters corresponding to the node type. It is a non-linear activation function. Encode the features of node v at time t. For node type The corresponding feature encoding mapping function is used to map the original features of node v at time t. Mapped to a latent space of a unified dimension.

[0091] The node encoding representation is input into the spatial dependency modeling part. This part constructs a multi-relation graph structure based on causal prior information (node ​​set, relation type, structure mask matrix, and edge weight prior information). Then, using a graph neural network, it models the spatial dependencies between nodes, obtaining the updated node spatial encoding representation. Preferably, a graph neural network with an attention mechanism can be used for spatial encoding, establishing propagation connections only between node pairs allowed by the structure mask, and weighting and aggregating the information propagation between nodes based on edge weight priors. For the ... Node representation in layered graph neural networks The weighted aggregation from adjacent nodes can be calculated as follows:

[0092] ;

[0093] in, For the spatial encoding of node v at time t in layer (l+1), It is a non-linear activation function. To represent the spatial encoding of node u at time t in layer l, This represents the set of neighboring nodes connected to node v at time t under relation type r; This is a linear transformation matrix corresponding to the relation type and the number of levels; These are the attention weights. By stacking several layers of the above graph neural network, node representations that fuse spatial dependency information at time t can be obtained. (i.e., node space encoding representation), where This indicates the number of layers in a graph neural network.

[0094] The spatial encoding representation of the nodes, which is the final output of the spatial dependency modeling part, is input into the temporal dependency modeling part according to a preset time window. For example, for node i, a time window of length L is selected, and the spatial encoding representation sequence from time step t-L+1 to t is used as the input for time series modeling. This temporal dependency modeling part can use a temporal convolutional network (TCN) for temporal modeling and output the comprehensive spatiotemporal representation of the nodes. for:

[0095] ;

[0096] in, Represents the spatial encoding of nodes. This indicates that a temporal convolution operation is being performed. The temporal convolution operation extracts node features within a window of time step t (from t-L+1 to t) to capture the temporal dependencies of nodes over the past L time steps, thereby updating the latent space representation of the nodes.

[0097] It should be noted that the time-dependent modeling part of the embodiments of this application can also use recurrent neural networks, long short-term memory networks (LSTM), gated recurrent units (GRU), or self-attention structures for time modeling.

[0098] The comprehensive spatiotemporal representation of the time-dependent modeling part is input into a fully connected network (e.g., a multilayer perceptron with 1-2 layers) to obtain the predicted value of the continuous pollution risk index at the target prediction time. :

[0099] ;

[0100] in, This represents the comprehensive spatiotemporal representation of node i at time t. The input is fed into a multilayer perceptron (MLP), which learns from multiple layers of fully connected networks and nonlinear activation functions. The nonlinear mapping to continuous pollution risk indicators outputs the target time. The continuous pollution risk prediction value, of which To predict the step size.

[0101] It should be noted that the spatiotemporal prediction model in this application embodiment can also introduce mechanisms such as random dropout to perform multiple forward calculations on the same input, thereby estimating the uncertainty of the prediction result.

[0102] The above spatiotemporal prediction model was trained using historical multi-source data and corresponding continuous pollution risk indicators to obtain an insulator pollution risk prediction model. During the training phase, for the training sample set... The mean squared error can be used as the basic prediction loss:

[0103] ;

[0104] in, To predict basic losses, For predicted values, Output the target value. and These are the node index and the time index, respectively.

[0105] With causal prior constraints, the attention weights of allowed edges under each relation type are averaged over time and layer to obtain the average attention weight. and normalized edge weight prior Comparison, constructing the loss of causal prior constraints , represented as:

[0106] ;

[0107] The above formula represents the deviation between the node attention weights learned by the computational model and the prior information of the edge weights. By adding this causal prior constraint term to the overall loss, the edge influence strength learned by the model is guided to maintain consistency with the causal prior.

[0108] Furthermore, to prevent overfitting, the model parameters can be adjusted by introducing... Regularization term, constructing the total loss function :

[0109] ;

[0110] in, and These are the weighting coefficients. For prediction model parameters The squared L2 norm is used for regularization. It is the sum of squares of the model parameters and is often used in the loss function as a constraint to prevent overfitting and enhance the model's generalization ability. The total loss function is minimized using stochastic gradient descent or adaptive learning rate optimization algorithms to obtain the trained model parameters, thus yielding the insulator pollution risk prediction model.

[0111] Furthermore, in step 5 of this application embodiment, the temporal feature vectors of various types of nodes are updated based on the newly collected multi-source data and preprocessed. The updated temporal feature vectors of various types of nodes are then input into the insulator pollution risk prediction model to obtain the continuous pollution risk prediction result.

[0112] Furthermore, the prediction method proposed in this application embodiment also includes:

[0113] The insulator pollution risk prediction model is optimized online, that is, the model parameters are optimized using data collected in actual application. The specific process is as described in the model training process above, and will not be repeated here.

[0114] The prediction method proposed in this application introduces mechanism-driven causal prior constraints based on the fusion of multi-source information, which improves the accuracy, stability and interpretability of insulator pollution risk prediction results, and provides reliable quantitative support for risk-based maintenance of outdoor insulators.

[0115] Based on the same technical concept described above, this application also proposes an insulator pollution risk assessment and prediction system, such as... Figure 3 As shown, the prediction system 300 includes:

[0116] The data acquisition and preprocessing unit 301 is configured to: acquire and preprocess multi-source data related to the pollution risk of outdoor insulators, and construct a spatiotemporally aligned multi-dimensional time-series feature vector; the multi-source data includes insulator status data, pollution source emission data, and meteorological environmental data. The specific multi-source data and its preprocessing methods are as described in step 1 above, and will not be repeated here.

[0117] The indicator construction unit 302 is configured to: construct a continuous pollution risk indicator for each target node based on a multi-dimensional temporal feature vector, serving as a label for the pollution risk intensity of the corresponding target node at each time point. The specific indicator construction method is as described in step 2 above, and will not be repeated here.

[0118] The causal prior construction unit 303 is configured to construct causal prior information to characterize the interaction between pollution source emissions, meteorological conditions, plant layout, and insulator location. The specific method for constructing causal prior information is as described in step 3 above and will not be repeated here.

[0119] The model training and update unit 304 is configured to: train a pre-established spatiotemporal prediction model using multi-dimensional temporal feature vectors and corresponding continuous pollution risk indicators to obtain an insulator pollution risk prediction model; this spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependencies between nodes under causal prior information constraints, while the temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk indicator at the prediction time. During training, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms. The specific model training process is as described in step 4 above and will not be repeated here.

[0120] Furthermore, the prediction unit 305 is configured to: acquire new multi-source data in real time using the data acquisition and preprocessing unit 301, perform preprocessing, update the multi-dimensional time-series feature vector, input the updated multi-dimensional time-series feature vector into the insulator pollution risk prediction model, and output the insulator pollution risk prediction result. The specific evaluation process is as described in step 5 above, and will not be repeated here.

[0121] Furthermore, the model training and update unit 304 in this embodiment is also configured as follows:

[0122] The multidimensional time-series feature vector is updated using the data acquisition and preprocessing unit 301, and the continuous pollution risk index is updated using the index construction unit 302. The model is then trained and updated using the updated multidimensional time-series feature vector and its corresponding continuous pollution risk index.

[0123] Based on the same technical concept described above, this application also proposes an electronic device, such as... Figure 4 As shown, the electronic device 400 includes: a memory 410, a processor 420, and a computer program A411 stored in the memory 410 and executable on the processor 420. When the processor 420 executes the computer program A411, it performs the following steps:

[0124] Multi-source data related to the pollution risk of outdoor insulators were collected and preprocessed to construct a spatiotemporally aligned multi-dimensional time-series feature vector. The multi-source data included insulator status data, pollution source emission data, and meteorological environmental data.

[0125] Based on multi-dimensional temporal feature vectors, a continuous pollution risk index is constructed for each target node, which serves as a label for the pollution risk intensity of the corresponding target node at each time.

[0126] Based on the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location, causal prior information is constructed to characterize the interaction.

[0127] A pre-established spatiotemporal prediction model is trained using multi-dimensional temporal feature vectors and corresponding continuous pollution risk indicators to obtain an insulator pollution risk prediction model. This spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of causal prior information, while the temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk indicator at the prediction time. During the training process, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms.

[0128] New multi-source data related to the pollution risk of outdoor insulators are collected in real time and preprocessed to update the multi-dimensional time-series feature vector. The updated multi-dimensional time-series feature vector is then input into the insulator pollution risk prediction model, and the insulator pollution risk prediction result is output.

[0129] Optionally, when the processor 420 executes the computer program A411, it may implement any of the embodiments in the corresponding examples of the prediction method described above.

[0130] It should be noted that the electronic device proposed in this application embodiment is a device used to implement the above prediction method. Therefore, based on the above prediction method proposed in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this application embodiment. Therefore, how the electronic device specifically implements the above prediction method will not be described in detail here. Any electronic device used by those skilled in the art to implement the above prediction method falls within the scope of protection of this application.

[0131] Based on the same technical concept described above, embodiments of this application also propose a computer-readable storage medium, such as... Figure 5 As shown, the computer-readable storage medium 500 stores a computer program B511, which, when executed by a processor, performs the following steps:

[0132] Multi-source data related to the pollution risk of outdoor insulators were collected and preprocessed to construct a spatiotemporally aligned multi-dimensional time-series feature vector. The multi-source data included insulator status data, pollution source emission data, and meteorological environmental data.

[0133] Based on multi-dimensional temporal feature vectors, a continuous pollution risk index is constructed for each target node, which serves as a label for the pollution risk intensity of the corresponding target node at each time.

[0134] Based on the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location, causal prior information is constructed to characterize the interaction.

[0135] A pre-established spatiotemporal prediction model is trained using multi-dimensional temporal feature vectors and corresponding continuous pollution risk indicators to obtain an insulator pollution risk prediction model. This spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of causal prior information, while the temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk indicator at the prediction time. During the training process, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms.

[0136] New multi-source data related to the pollution risk of outdoor insulators are collected in real time and preprocessed to update the multi-dimensional time-series feature vector. The updated multi-dimensional time-series feature vector is then input into the insulator pollution risk prediction model, and the insulator pollution risk prediction result is output.

[0137] Optionally, when the computer program B511 is executed by the processor, it can implement any of the embodiments corresponding to the above prediction method.

[0138] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0139] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0140] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0141] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0142] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0143] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for assessing and predicting pollution risk in insulators, characterized in that, include: Multi-source data related to the pollution risk of outdoor insulators were collected and preprocessed to construct a spatiotemporally aligned multidimensional temporal feature vector. The multi-source data includes insulator status data, pollution source emission data, and meteorological environmental data. Based on the multidimensional temporal feature vector, a continuous pollution risk index is constructed for each target node, serving as a label for the pollution risk intensity of the corresponding target node at each time. Based on the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location, causal prior information is constructed to characterize the interaction. The pre-established spatiotemporal prediction model is trained using the multi-dimensional temporal feature vector and the corresponding continuous pollution risk index to obtain the insulator pollution risk prediction model. The spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part. The spatial dependency modeling part models the spatial dependency between nodes under the constraint of the causal prior information. The temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk index at the prediction time. During training, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms. New multi-source data related to the pollution risk of outdoor insulators are collected in real time and preprocessed to update the multi-dimensional time-series feature vector. The updated multi-dimensional time-series feature vector is then input into the insulator pollution risk prediction model, and the insulator pollution risk prediction result is output. The process of constructing the causal prior information includes: Constructing node types and node sets: Abstracting relevant entities in the scene into nodes of different types to form node sets, and associating each node in the node set with its node type, including substation / insulator node type, pollution source node type and meteorological environment node type; Constructing Relationship Types: Based on physical mechanisms and engineering experience, several types of relationships between nodes are defined to form a relationship set. The relationship set includes pollutant transport relationships, environmental similarity relationships, and plant topology or electrical connection relationships. Construct a structure mask matrix: For each relation type in the relation set, construct a structure mask matrix. The elements in the structure mask matrix are used to indicate whether a directed edge from one node to another is allowed under the corresponding relation type. If it exists, the element is set to 1; otherwise, it is set to 0. Constructing edge weight prior information: Under the premise of structural masking, edge weight prior is defined based on the distance between nodes, wind direction and speed, emission intensity and meteorological similarity. This prior is used to characterize the relative influence of one node on another node in the corresponding relationship type, and the edge weight prior is normalized. The construction and training process of the insulator pollution risk prediction model includes: The multidimensional temporal feature vector is input into the node feature encoding part, which maps the temporal feature vectors of the input heterogeneous nodes to a latent space of a unified dimension and outputs the node encoded representation to the spatial dependency modeling part. The spatial dependency modeling part constructs a multi-relationship graph structure based on the causal prior information, and models the spatial dependency between nodes through a graph neural network to obtain an updated node spatial encoding representation. The node spatial encoding representation is input into the time dependency modeling part according to a preset time window. The time dependency modeling part performs time modeling and outputs the comprehensive spatiotemporal representation of the node. By inputting the comprehensive spatiotemporal representation of the nodes into the fully connected network, the predicted value of the continuous pollution risk index at the target prediction time is obtained.

2. The method for assessing and predicting insulator pollution risk according to claim 1, characterized in that, The multidimensional time-series feature vector includes: long-term pollution accumulation features, short-term excitation features, and pollution source-related features; The long-term pollution accumulation feature represents the degree of long-term pollution accumulation and is extracted from the insulator condition data. The short-term excitation feature represents the characteristics of short-term excitation conditions and is extracted from the meteorological environment data. The pollution source characteristics represent the characteristics of the impact of upstream pollution sources, which are extracted from the pollution source emission data.

3. The insulator pollution risk assessment and prediction method according to claim 2, characterized in that, The continuous pollution risk indicator is constructed as follows: By weighting and normalizing the long-term pollution accumulation characteristics, short-term excitation characteristics, and pollution source-related characteristics, a continuous pollution risk index is obtained.

4. A method for assessing and predicting insulator pollution risk according to any one of claims 1-3, characterized in that, The spatial dependency modeling part uses a graph neural network with an attention mechanism for spatial encoding. It establishes propagation connections only between node pairs allowed by the structural mask and performs weighted aggregation of information propagation between nodes based on edge weight priors. Through several layers of graph neural networks, a node spatial encoding representation that integrates spatial dependency information is obtained.

5. A method for assessing and predicting insulator pollution risk according to any one of claims 1-3, characterized in that, The constraint term is constructed by calculating the deviation between the attention weights between nodes learned by the model and the prior information of edge weights, and using this deviation as part of the total loss function to guide the edge influence strength learned by the model to be consistent with the causal prior information.

6. A system for assessing and predicting pollution risk in insulators, characterized in that, include: The data acquisition and preprocessing unit is configured to: acquire multi-source data related to the pollution risk of outdoor insulators and preprocess it to construct a spatiotemporally aligned multi-dimensional time-series feature vector; the multi-source data includes insulator status data, pollution source emission data and meteorological environment data. The indicator construction unit is configured to: construct a continuous pollution risk indicator for each target node based on the multi-dimensional time-series feature vector, which serves as a label for the pollution risk intensity of the corresponding target node at each time. The causal prior construction unit is configured to: construct causal prior information to characterize the interaction between pollution source emissions, meteorological conditions, plant layout and insulator location; The model training and update unit is configured to: train a pre-established spatiotemporal prediction model using the multi-dimensional temporal feature vector and the corresponding continuous pollution risk index to obtain an insulator pollution risk prediction model; wherein the spatiotemporal prediction model includes a node feature encoding part, a spatial dependency modeling part, and a temporal dependency modeling part, wherein the spatial dependency modeling part models the spatial dependency between nodes under the constraint of the causal prior information, and the temporal dependency modeling part models the spatial feature sequence of the target node within a preset time window to output the predicted value of the continuous pollution risk index at the prediction time; During training, constraint terms and parameter regularization terms based on causal prior information are introduced to optimize the loss function, which includes prediction error terms, constraint terms, and parameter regularization terms. And, the prediction unit is configured to: use the data acquisition and preprocessing unit to acquire new multi-source data in real time and preprocess it, update the multi-dimensional time series feature vector, input the updated multi-dimensional time series feature vector into the insulator pollution risk prediction model, and output the insulator pollution risk prediction result. The process of constructing the causal prior information includes: Constructing node types and node sets: Abstracting relevant entities in the scene into nodes of different types to form node sets, and associating each node in the node set with its node type, including substation / insulator node type, pollution source node type and meteorological environment node type; Constructing Relationship Types: Based on physical mechanisms and engineering experience, several types of relationships between nodes are defined to form a relationship set. The relationship set includes pollutant transport relationships, environmental similarity relationships, and plant topology or electrical connection relationships. Construct a structure mask matrix: For each relation type in the relation set, construct a structure mask matrix. The elements in the structure mask matrix are used to indicate whether a directed edge from one node to another is allowed under the corresponding relation type. If it exists, the element is set to 1; otherwise, it is set to 0. Constructing edge weight prior information: Under the premise of structural masking, edge weight prior is defined based on the distance between nodes, wind direction and speed, emission intensity and meteorological similarity. This prior is used to characterize the relative influence of one node on another node in the corresponding relationship type, and the edge weight prior is normalized. The construction and training process of the insulator pollution risk prediction model includes: The multidimensional temporal feature vector is input into the node feature encoding part, which maps the temporal feature vectors of the input heterogeneous nodes to a latent space of a unified dimension and outputs the node encoded representation to the spatial dependency modeling part. The spatial dependency modeling part constructs a multi-relationship graph structure based on the causal prior information, and models the spatial dependency between nodes through a graph neural network to obtain an updated node spatial encoding representation. The node spatial encoding representation is input into the time dependency modeling part according to a preset time window. The time dependency modeling part performs time modeling and outputs the comprehensive spatiotemporal representation of the node. By inputting the comprehensive spatiotemporal representation of the nodes into the fully connected network, the predicted value of the continuous pollution risk index at the target prediction time is obtained.

7. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the insulator pollution risk assessment and prediction method according to any one of claims 1-5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the insulator pollution risk assessment and prediction method according to any one of claims 1-5.

Images